MemGovern: Enhancing Code Agents through Learning from Governed Human Experiences
Qihao Wang, Ziming Cheng, Shuo Zhang, Fan Liu, Rui Xu, Heng Lian, Kunyi Wang, Xiaoming Yu, Jianghao Yin, Sen Hu, Yue Hu, Shaolei Zhang, Yanbing Liu, Ronghao Chen, Huacan Wang
2026-01-14
Summary
This paper introduces a new system called MemGovern that helps AI programming tools, known as autonomous software engineering (SWE) agents, learn from past human solutions to programming problems found on platforms like GitHub.
What's the problem?
Currently, these AI agents are limited because they try to solve problems on their own or only look at the immediate code around a bug, ignoring the vast amount of knowledge available in places like GitHub where programmers have already fixed similar issues. The information on GitHub is messy and hard for the AI to understand and use effectively, it's not organized in a way that's helpful for them.
What's the solution?
The researchers created MemGovern, which acts like a librarian for GitHub. It takes all the raw data from GitHub issue trackers and transforms it into organized 'experience cards' that the AI can easily understand. It also developed a way for the AI to specifically search for and retrieve relevant past solutions based on the logic of the problem. They created over 135,000 of these experience cards and tested it on a standard programming benchmark.
Why it matters?
This work is important because it allows AI programming tools to learn from the collective experience of human programmers, making them much more effective at fixing bugs and writing code. By giving these agents access to this 'open-world' knowledge, it significantly improves their performance, in this case increasing bug resolution rates by almost 4.7%, and provides a way to easily add this memory capability to existing AI tools.
Abstract
While autonomous software engineering (SWE) agents are reshaping programming paradigms, they currently suffer from a "closed-world" limitation: they attempt to fix bugs from scratch or solely using local context, ignoring the immense historical human experience available on platforms like GitHub. Accessing this open-world experience is hindered by the unstructured and fragmented nature of real-world issue-tracking data. In this paper, we introduce MemGovern, a framework designed to govern and transform raw GitHub data into actionable experiential memory for agents. MemGovern employs experience governance to convert human experience into agent-friendly experience cards and introduces an agentic experience search strategy that enables logic-driven retrieval of human expertise. By producing 135K governed experience cards, MemGovern achieves a significant performance boost, improving resolution rates on the SWE-bench Verified by 4.65%. As a plug-in approach, MemGovern provides a solution for agent-friendly memory infrastructure.